libretro/Lakka-LibreELEC
0
0
Fork 1
mirror of https://github.com/libretro/Lakka-LibreELEC.git synced 2025-04-12 12:53:10 +00:00
Code Issues Activity
Files
Lakka-LibreELEC/projects/RPi/patches/mesa/0002-backport-2712D0-support.patch
Tomáš Kelemen d6bec8fcce revert commits from LibreELEC master / not needed commits (#2039) 
this brings the branch to a state, where vulkan works

Revert "mesa: update to 24.3.4" (75f30be9e6713f5491b68f5e03aeee1668cae7b8)
Revert "ply: add python-ply package" (51611919f25a9655be826182b7117eefae53215d)
Revert "wayland: fix PKG_DEPENDS_HOST" (5102cb8eee9e497e9801bee41fbee66046e2fdf9)
Revert "wayland: update to 1.23.1" (f96ac43c3beec40a92d9524878bad0ba2f10b9b4)
Revert "wayland: update to 1.23.0" (8957b5c8ba404dedadca8ef8e7555609b40e8933)
Revert "wayland-protocols: update to 1.39" (1a1531bbb0f876fbc4c67624a21b950b6ca3c55c)
Revert "Update spirv-headers spirv-llvm-translator spirv-tools glslang llvm packages" (e907f3cfdf2d6a8f11729d9b59ea9c11f184927d)
Revert "kmod: update to 33" (1d4adfcf3cafc279749f6ac33a19e5db81ef0151)
Revert "LLVM 19 (#2019)" (3d815fe80b45193c91997b92cdf194326aa4c80a)
Revert "Rust 1.83 (#2020)" (501a533c129455cab34d9b1a70ad8cd092c16828)
Revert "spirv-tools: move package.mk from vulkan to graphics" (71cb84343a0bb27df0fce69f21db146d24dc390e)
Revert "spirv-headers: move package.mk from vulkan to graphics" (49bf112b3087a93f765908a4a8e2e38f3673b05d)
Revert "pycparser: initial package" (cfbf6d2119f1a8f8eb2488447d3c8af74a3204ec)
Revert "glslang: update unpack" (d4641dc629f0e03aa57ee94f8426a30244fb4007)
Revert "volk: update to 1.3.295" (7ef40616d0e9c14f66d9ece18c6fba4665a04da6)
Revert "llvm: include build of llvm-spirv" (128aac7324574b4197ff6915351953ccdf828c57)
Revert "json-c: fix package url" (6254ec9c17568ae4828a6cc2668410609745dd5b)
Revert "xxHash: fix toolchain target directory" (2bde0c9df99de4b98d52a1aa3c6fbe5335b1ee25)
Revert "xxHash: initial package" (17059b0e404d9d6b7a5f5e9c57a2fe872886831f)
Revert "Remove kmsro references" (22d233fa1da867f4bc4d1287871ac3b94acb4172)
Revert "pyyaml: initial package" (09cb87fcd765b79093ee19cf738b5e16aa722cc8)
Revert "mesa: clean no-op" (5a63fed4d1ab20b8b4f5e35090c2549438cb7f8e)
Revert "libclc: initial package" (83436943f96ec9f6efb23125f7f43c1dbde42039)
Revert "libdrm: update to 2.4.124" (efbc571824ee0a9553fbfd5971d3d6f7bcc456c5)
Revert "libdrm: update to 2.4.123" (ae2cecd5c32a0be1fe9b3a0ecbc88caf35caea1e)
Revert "cleanup packages: various" (4d0594607ebfa42d7ca5d1d96d8d6684ca9dcd20)
Revert "libdrm: update to 2.4.122" (8f69530bf70cac1e7a2dc6f997a842d4353143a5)
Revert "libdrm: update to 2.4.121" (b693dfd8656114b9da18c57d8e72366799c895de)
Revert "libdrm: enable host build" (6b3137c54cf53c757c22accac3c13bd5e99a66b2)
Revert "spirv-llvm-translator: update to 18.1.4" (b147453542135fe0d267cc99c4260af25003ada2)
Revert "spirv-llvm-translator: update to 18.1.3" (56ae7902eeaa469a04b3f73ee85d5985e826b0c3)
Revert "spirv-llvm-translator: update to 18.1.2" (633a7dace8972ee8c4cc9459cf51052bfa327a34)
Revert "spirv-llvm-translator: update to 18.1.1" (5b44164ee57c1deeaaebe27a794e0d7d65e0f332)
Revert "spirv-llvm-translator: initial package" (b2062cb474c17f16527deab176fd0a84c2fe67b0)
Revert "mesa: update to 24.3.1" (719c646240cbfaba40dd7b569ab3998833706cc1)
Revert "Mesa: Add missing bracket" (a90097eade991ff730e31cf9c37095fc141d55b3)
Revert "Fix libxv package" (6a895e4e6a32ca07b4454c924f160a51f7549ada)
Revert "mesa: update to 24.3.0" (d24ac0c0867b3d1ce4babc503e17533e6c4e8656)
Revert "mesa v24.2.7" (721e92ae82bf82cb34b488674fb7e3a7d13260ac)
Revert "mesa: update to 24.2.4" (e551d0cf1d685e51f13c63a2eae8e9391c3390c4)
Revert "mesa: update to 24.2.3" (b9e314a054f3c5f4dcc7d5012e030e0ebcfbf1b1)
Revert "mesa: update to 24.2.2" (bc9c2d06331a187fc4863a211133b5b7599f6d5a)
Revert "mesa: update to 24.2.1" (0c731fb694079c5ee9bc4a5efa1c795f3df72fe7)
Revert "mesa: update to 24.2.0" (6f8a7e2e60912b93437a7efa69a6cff2d81bb3bd)
Revert "mesa v24.1.5" (211ce11e84e7d7cdbdbbcae59ee646e6c529169c)
Revert "mesa: update to 24.1.0" (e66319dea5d86a460ea6070c9512fdf53ac72b20)
Revert "Eliminate wiringpi 404" (f6a989033e01d4e4f35e02a6c15307e2535bdf3a)
2025-02-20 17:52:25 +01:00

714 lines
36 KiB
Diff
Raw Permalink Blame History

From 47e68232c04cd35fc9361cffcafdc76cb99a76eb Mon Sep 17 00:00:00 2001
From: Iago Toral Quiroga <itoral@igalia.com>
Date: Tue, 14 May 2024 09:40:16 +0200
Subject: [PATCH 1/3] broadcom/cle: fix up shader record for V3D 7.1.10 /
2712D0
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29189>
---
src/broadcom/cle/v3d_packet.xml | 63 +++++++++++++++++++++++++++++++++
1 file changed, 63 insertions(+)
diff --git a/src/broadcom/cle/v3d_packet.xml b/src/broadcom/cle/v3d_packet.xml
index 09dde392fac3..4763b891aab1 100644
--- a/src/broadcom/cle/v3d_packet.xml
+++ b/src/broadcom/cle/v3d_packet.xml
@@ -1211,6 +1211,69 @@
<field name="Coordinate Shader Uniforms Address" size="32" start="28b" type="address"/>
</struct>
+ <!-- This one is exclusive to 2712D0 (V3D 7.1.10) -->
+ <struct name="GL Shader State Record Draw Index" min_ver="71">
+ <field name="Point size in shaded vertex data" size="1" start="0" type="bool"/>
+ <field name="Enable clipping" size="1" start="1" type="bool"/>
+
+ <field name="Vertex ID read by coordinate shader" size="1" start="2" type="bool"/>
+ <field name="Instance ID read by coordinate shader" size="1" start="3" type="bool"/>
+ <field name="Base Instance ID read by coordinate shader" size="1" start="4" type="bool"/>
+ <field name="cs_basevertex" size="1" start="5" type="bool"/>
+ <field name="cs_drawindex" size="1" start="6" type="bool"/>
+
+ <field name="Vertex ID read by vertex shader" size="1" start="7" type="bool"/>
+ <field name="Instance ID read by vertex shader" size="1" start="8" type="bool"/>
+ <field name="Base Instance ID read by vertex shader" size="1" start="9" type="bool"/>
+ <field name="vs_basevertex" size="1" start="10" type="bool"/>
+ <field name="vs_drawindex" size="1" start="11" type="bool"/>
+
+ <field name="Fragment shader does Z writes" size="1" start="12" type="bool"/>
+ <field name="Turn off early-z test" size="1" start="13" type="bool"/>
+
+ <field name="Fragment shader uses real pixel centre W in addition to centroid W2" size="1" start="15" type="bool"/>
+ <field name="Enable Sample Rate Shading" size="1" start="16" type="bool"/>
+ <field name="Any shader reads hardware-written Primitive ID" size="1" start="17" type="bool"/>
+ <field name="Insert Primitive ID as first varying to fragment shader" size="1" start="18" type="bool"/>
+ <field name="Turn off scoreboard" size="1" start="19" type="bool"/>
+ <field name="Do scoreboard wait on first thread switch" size="1" start="20" type="bool"/>
+ <field name="Disable implicit point/line varyings" size="1" start="21" type="bool"/>
+ <field name="No prim pack" size="1" start="22" type="bool"/>
+ <field name="Never defer FEP depth writes" size="1" start="23" type="bool"/>
+
+ <field name="Number of varyings in Fragment Shader" size="8" start="3b" type="uint"/>
+
+ <field name="Coordinate Shader output VPM segment size" size="4" start="4b" type="uint"/>
+ <field name="Min Coord Shader output segments required in play in addition to VCM cache size" size="4" start="36" type="uint"/>
+
+ <field name="Coordinate Shader input VPM segment size" size="4" start="5b" type="uint"/>
+ <field name="Min Coord Shader input segments required in play" size="4" start="44" type="uint" minus_one="true"/>
+
+ <field name="Vertex Shader output VPM segment size" size="4" start="6b" type="uint"/>
+ <field name="Min Vertex Shader output segments required in play in addition to VCM cache size" size="4" start="52" type="uint"/>
+
+ <field name="Vertex Shader input VPM segment size" size="4" start="7b" type="uint"/>
+ <field name="Min Vertex Shader input segments required in play" size="4" start="60" type="uint" minus_one="true"/>
+
+ <field name="Fragment Shader Code Address" size="29" start="67" type="address"/>
+ <field name="Fragment Shader 4-way threadable" size="1" start="64" type="bool"/>
+ <field name="Fragment Shader start in final thread section" size="1" start="65" type="bool"/>
+ <field name="Fragment Shader Propagate NaNs" size="1" start="66" type="bool"/>
+ <field name="Fragment Shader Uniforms Address" size="32" start="12b" type="address"/>
+
+ <field name="Vertex Shader Code Address" size="29" start="131" type="address"/>
+ <field name="Vertex Shader 4-way threadable" size="1" start="128" type="bool"/>
+ <field name="Vertex Shader start in final thread section" size="1" start="129" type="bool"/>
+ <field name="Vertex Shader Propagate NaNs" size="1" start="130" type="bool"/>
+ <field name="Vertex Shader Uniforms Address" size="32" start="20b" type="address"/>
+
+ <field name="Coordinate Shader Code Address" size="29" start="195" type="address"/>
+ <field name="Coordinate Shader 4-way threadable" size="1" start="192" type="bool"/>
+ <field name="Coordinate Shader start in final thread section" size="1" start="193" type="bool"/>
+ <field name="Coordinate Shader Propagate NaNs" size="1" start="194" type="bool"/>
+ <field name="Coordinate Shader Uniforms Address" size="32" start="28b" type="address"/>
+ </struct>
+
<struct name="Geometry Shader State Record">
<field name="Geometry Bin Mode Shader Code Address" size="29" start="3" type="address"/>
<field name="Geometry Bin Mode Shader 4-way threadable" size="1" start="0" type="bool"/>
--
2.39.2
From 21153810ee2c6b3a3a8943ba442f473ee977d5ff Mon Sep 17 00:00:00 2001
From: Iago Toral Quiroga <itoral@igalia.com>
Date: Tue, 14 May 2024 10:24:07 +0200
Subject: [PATCH 2/3] v3d: support 2712D0
2710D0 has V3D 7.1.10 which included draw index and
base vertex in the shader state record packet, shuffling
the locations of most of its fields. Handle this at run
time by emitting the appropriate packet based on the
V3D version since our current versoning framework doesn't
support changes based on revision number alone.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29189>
---
src/broadcom/common/v3d_device_info.h | 6 +
src/gallium/drivers/v3d/v3dx_draw.c | 316 +++++++++++++++++---------
2 files changed, 214 insertions(+), 108 deletions(-)
diff --git a/src/broadcom/common/v3d_device_info.h b/src/broadcom/common/v3d_device_info.h
index 8dfc7858727e..0df25e8dcce3 100644
--- a/src/broadcom/common/v3d_device_info.h
+++ b/src/broadcom/common/v3d_device_info.h
@@ -52,4 +52,10 @@ typedef int (*v3d_ioctl_fun)(int fd, unsigned long request, void *arg);
bool
v3d_get_device_info(int fd, struct v3d_device_info* devinfo, v3d_ioctl_fun fun);
+static inline bool
+v3d_device_has_draw_index(struct v3d_device_info *devinfo)
+{
+ return devinfo->ver > 71 || (devinfo->ver == 71 && devinfo->rev >= 10);
+}
+
#endif
diff --git a/src/gallium/drivers/v3d/v3dx_draw.c b/src/gallium/drivers/v3d/v3dx_draw.c
index 128104c213b7..85922f36c4e6 100644
--- a/src/gallium/drivers/v3d/v3dx_draw.c
+++ b/src/gallium/drivers/v3d/v3dx_draw.c
@@ -472,114 +472,100 @@ v3d_emit_tes_gs_shader_params(struct v3d_job *job,
}
static void
-v3d_emit_gl_shader_state(struct v3d_context *v3d,
- const struct pipe_draw_info *info)
+emit_shader_state_record(struct v3d_context *v3d,
+ struct v3d_job *job,
+ const struct pipe_draw_info *info,
+ struct v3d_vertex_stateobj *vtx,
+ struct v3d_cl_reloc cs_uniforms,
+ struct v3d_cl_reloc vs_uniforms,
+ struct v3d_cl_reloc fs_uniforms,
+ struct vpm_config *vpm_cfg_bin,
+ struct vpm_config *vpm_cfg)
{
- struct v3d_job *job = v3d->job;
- /* V3D_DIRTY_VTXSTATE */
- struct v3d_vertex_stateobj *vtx = v3d->vtx;
- /* V3D_DIRTY_VTXBUF */
- struct v3d_vertexbuf_stateobj *vertexbuf = &v3d->vertexbuf;
-
- /* Upload the uniforms to the indirect CL first */
- struct v3d_cl_reloc fs_uniforms =
- v3d_write_uniforms(v3d, job, v3d->prog.fs,
- PIPE_SHADER_FRAGMENT);
-
- struct v3d_cl_reloc gs_uniforms = { NULL, 0 };
- struct v3d_cl_reloc gs_bin_uniforms = { NULL, 0 };
- if (v3d->prog.gs) {
- gs_uniforms = v3d_write_uniforms(v3d, job, v3d->prog.gs,
- PIPE_SHADER_GEOMETRY);
- }
- if (v3d->prog.gs_bin) {
- gs_bin_uniforms = v3d_write_uniforms(v3d, job, v3d->prog.gs_bin,
- PIPE_SHADER_GEOMETRY);
- }
-
- struct v3d_cl_reloc vs_uniforms =
- v3d_write_uniforms(v3d, job, v3d->prog.vs,
- PIPE_SHADER_VERTEX);
- struct v3d_cl_reloc cs_uniforms =
- v3d_write_uniforms(v3d, job, v3d->prog.cs,
- PIPE_SHADER_VERTEX);
-
- /* Update the cache dirty flag based on the shader progs data */
- job->tmu_dirty_rcl |= v3d->prog.cs->prog_data.vs->base.tmu_dirty_rcl;
- job->tmu_dirty_rcl |= v3d->prog.vs->prog_data.vs->base.tmu_dirty_rcl;
- if (v3d->prog.gs_bin) {
- job->tmu_dirty_rcl |=
- v3d->prog.gs_bin->prog_data.gs->base.tmu_dirty_rcl;
- }
- if (v3d->prog.gs) {
- job->tmu_dirty_rcl |=
- v3d->prog.gs->prog_data.gs->base.tmu_dirty_rcl;
- }
- job->tmu_dirty_rcl |= v3d->prog.fs->prog_data.fs->base.tmu_dirty_rcl;
-
- uint32_t num_elements_to_emit = 0;
- for (int i = 0; i < vtx->num_elements; i++) {
- struct pipe_vertex_element *elem = &vtx->pipe[i];
- struct pipe_vertex_buffer *vb =
- &vertexbuf->vb[elem->vertex_buffer_index];
- if (vb->buffer.resource)
- num_elements_to_emit++;
- }
-
- uint32_t shader_state_record_length =
- cl_packet_length(GL_SHADER_STATE_RECORD);
- if (v3d->prog.gs) {
- shader_state_record_length +=
- cl_packet_length(GEOMETRY_SHADER_STATE_RECORD) +
- cl_packet_length(TESSELLATION_GEOMETRY_COMMON_PARAMS) +
- 2 * cl_packet_length(TESSELLATION_GEOMETRY_SHADER_PARAMS);
- }
-
- /* See GFXH-930 workaround below */
- uint32_t shader_rec_offset =
- v3d_cl_ensure_space(&job->indirect,
- shader_state_record_length +
- MAX2(num_elements_to_emit, 1) *
- cl_packet_length(GL_SHADER_STATE_ATTRIBUTE_RECORD),
- 32);
-
- /* XXX perf: We should move most of the SHADER_STATE_RECORD setup to
- * compile time, so that we mostly just have to OR the VS and FS
- * records together at draw time.
+#if V3D_VERSION >= 71
+ /* 2712D0 (V3D 7.1.10) has included draw index and base vertex,
+ * shuffling all the fields in the packet. Since the versioning
+ * framework doesn't handle revision numbers, the XML has a
+ * different shader state record packet including the new fields
+ * and we decide at run time which packet we need to emit.
*/
-
- struct vpm_config vpm_cfg_bin, vpm_cfg;
- v3d_compute_vpm_config(&v3d->screen->devinfo,
- v3d->prog.cs->prog_data.vs,
- v3d->prog.vs->prog_data.vs,
- v3d->prog.gs ? v3d->prog.gs_bin->prog_data.gs : NULL,
- v3d->prog.gs ? v3d->prog.gs->prog_data.gs : NULL,
- &vpm_cfg_bin,
- &vpm_cfg);
-
- if (v3d->prog.gs) {
- v3d_emit_gs_state_record(v3d->job,
- v3d->prog.gs_bin, gs_bin_uniforms,
- v3d->prog.gs, gs_uniforms);
-
- struct v3d_gs_prog_data *gs = v3d->prog.gs->prog_data.gs;
- v3d_emit_tes_gs_common_params(v3d->job,
- gs->out_prim_type,
- gs->num_invocations);
-
- /* Bin Tes/Gs params */
- v3d_emit_tes_gs_shader_params(v3d->job,
- vpm_cfg_bin.gs_width,
- vpm_cfg_bin.Gd,
- vpm_cfg_bin.Gv);
-
- /* Render Tes/Gs params */
- v3d_emit_tes_gs_shader_params(v3d->job,
- vpm_cfg.gs_width,
- vpm_cfg.Gd,
- vpm_cfg.Gv);
+ if (v3d_device_has_draw_index(&v3d->screen->devinfo)) {
+ cl_emit(&job->indirect, GL_SHADER_STATE_RECORD_DRAW_INDEX, shader) {
+ shader.enable_clipping = true;
+ shader.point_size_in_shaded_vertex_data =
+ (info->mode == MESA_PRIM_POINTS &&
+ v3d->rasterizer->base.point_size_per_vertex);
+ shader.fragment_shader_does_z_writes =
+ v3d->prog.fs->prog_data.fs->writes_z;
+ shader.turn_off_early_z_test =
+ v3d->prog.fs->prog_data.fs->disable_ez;
+ shader.fragment_shader_uses_real_pixel_centre_w_in_addition_to_centroid_w2 =
+ v3d->prog.fs->prog_data.fs->uses_center_w;
+ shader.any_shader_reads_hardware_written_primitive_id =
+ (v3d->prog.gs && v3d->prog.gs->prog_data.gs->uses_pid) ||
+ v3d->prog.fs->prog_data.fs->uses_pid;
+ shader.insert_primitive_id_as_first_varying_to_fragment_shader =
+ !v3d->prog.gs && v3d->prog.fs->prog_data.fs->uses_pid;
+ shader.do_scoreboard_wait_on_first_thread_switch =
+ v3d->prog.fs->prog_data.fs->lock_scoreboard_on_first_thrsw;
+ shader.disable_implicit_point_line_varyings =
+ !v3d->prog.fs->prog_data.fs->uses_implicit_point_line_varyings;
+ shader.number_of_varyings_in_fragment_shader =
+ v3d->prog.fs->prog_data.fs->num_inputs;
+ shader.coordinate_shader_code_address =
+ cl_address(v3d_resource(v3d->prog.cs->resource)->bo,
+ v3d->prog.cs->offset);
+ shader.vertex_shader_code_address =
+ cl_address(v3d_resource(v3d->prog.vs->resource)->bo,
+ v3d->prog.vs->offset);
+ shader.fragment_shader_code_address =
+ cl_address(v3d_resource(v3d->prog.fs->resource)->bo,
+ v3d->prog.fs->offset);
+ shader.coordinate_shader_input_vpm_segment_size =
+ v3d->prog.cs->prog_data.vs->vpm_input_size;
+ shader.vertex_shader_input_vpm_segment_size =
+ v3d->prog.vs->prog_data.vs->vpm_input_size;
+ shader.coordinate_shader_output_vpm_segment_size =
+ v3d->prog.cs->prog_data.vs->vpm_output_size;
+ shader.vertex_shader_output_vpm_segment_size =
+ v3d->prog.vs->prog_data.vs->vpm_output_size;
+ shader.coordinate_shader_uniforms_address = cs_uniforms;
+ shader.vertex_shader_uniforms_address = vs_uniforms;
+ shader.fragment_shader_uniforms_address = fs_uniforms;
+ shader.min_coord_shader_input_segments_required_in_play =
+ vpm_cfg_bin->As;
+ shader.min_vertex_shader_input_segments_required_in_play =
+ vpm_cfg->As;
+ shader.min_coord_shader_output_segments_required_in_play_in_addition_to_vcm_cache_size =
+ vpm_cfg_bin->Ve;
+ shader.min_vertex_shader_output_segments_required_in_play_in_addition_to_vcm_cache_size =
+ vpm_cfg->Ve;
+ shader.coordinate_shader_4_way_threadable =
+ v3d->prog.cs->prog_data.vs->base.threads == 4;
+ shader.vertex_shader_4_way_threadable =
+ v3d->prog.vs->prog_data.vs->base.threads == 4;
+ shader.fragment_shader_4_way_threadable =
+ v3d->prog.fs->prog_data.fs->base.threads == 4;
+ shader.coordinate_shader_start_in_final_thread_section =
+ v3d->prog.cs->prog_data.vs->base.single_seg;
+ shader.vertex_shader_start_in_final_thread_section =
+ v3d->prog.vs->prog_data.vs->base.single_seg;
+ shader.fragment_shader_start_in_final_thread_section =
+ v3d->prog.fs->prog_data.fs->base.single_seg;
+ shader.vertex_id_read_by_coordinate_shader =
+ v3d->prog.cs->prog_data.vs->uses_vid;
+ shader.instance_id_read_by_coordinate_shader =
+ v3d->prog.cs->prog_data.vs->uses_iid;
+ shader.vertex_id_read_by_vertex_shader =
+ v3d->prog.vs->prog_data.vs->uses_vid;
+ shader.instance_id_read_by_vertex_shader =
+ v3d->prog.vs->prog_data.vs->uses_iid;
+ }
+ return;
}
+#endif
+ assert(!v3d_device_has_draw_index(&v3d->screen->devinfo));
cl_emit(&job->indirect, GL_SHADER_STATE_RECORD, shader) {
shader.enable_clipping = true;
/* V3D_DIRTY_PRIM_MODE | V3D_DIRTY_RASTERIZER */
@@ -669,14 +655,14 @@ v3d_emit_gl_shader_state(struct v3d_context *v3d,
shader.fragment_shader_uniforms_address = fs_uniforms;
shader.min_coord_shader_input_segments_required_in_play =
- vpm_cfg_bin.As;
+ vpm_cfg_bin->As;
shader.min_vertex_shader_input_segments_required_in_play =
- vpm_cfg.As;
+ vpm_cfg->As;
shader.min_coord_shader_output_segments_required_in_play_in_addition_to_vcm_cache_size =
- vpm_cfg_bin.Ve;
+ vpm_cfg_bin->Ve;
shader.min_vertex_shader_output_segments_required_in_play_in_addition_to_vcm_cache_size =
- vpm_cfg.Ve;
+ vpm_cfg->Ve;
shader.coordinate_shader_4_way_threadable =
v3d->prog.cs->prog_data.vs->base.threads == 4;
@@ -707,6 +693,120 @@ v3d_emit_gl_shader_state(struct v3d_context *v3d,
vtx->defaults_offset);
#endif
}
+}
+
+static void
+v3d_emit_gl_shader_state(struct v3d_context *v3d,
+ const struct pipe_draw_info *info)
+{
+ struct v3d_job *job = v3d->job;
+ /* V3D_DIRTY_VTXSTATE */
+ struct v3d_vertex_stateobj *vtx = v3d->vtx;
+ /* V3D_DIRTY_VTXBUF */
+ struct v3d_vertexbuf_stateobj *vertexbuf = &v3d->vertexbuf;
+
+ /* Upload the uniforms to the indirect CL first */
+ struct v3d_cl_reloc fs_uniforms =
+ v3d_write_uniforms(v3d, job, v3d->prog.fs,
+ PIPE_SHADER_FRAGMENT);
+
+ struct v3d_cl_reloc gs_uniforms = { NULL, 0 };
+ struct v3d_cl_reloc gs_bin_uniforms = { NULL, 0 };
+ if (v3d->prog.gs) {
+ gs_uniforms = v3d_write_uniforms(v3d, job, v3d->prog.gs,
+ PIPE_SHADER_GEOMETRY);
+ }
+ if (v3d->prog.gs_bin) {
+ gs_bin_uniforms = v3d_write_uniforms(v3d, job, v3d->prog.gs_bin,
+ PIPE_SHADER_GEOMETRY);
+ }
+
+ struct v3d_cl_reloc vs_uniforms =
+ v3d_write_uniforms(v3d, job, v3d->prog.vs,
+ PIPE_SHADER_VERTEX);
+ struct v3d_cl_reloc cs_uniforms =
+ v3d_write_uniforms(v3d, job, v3d->prog.cs,
+ PIPE_SHADER_VERTEX);
+
+ /* Update the cache dirty flag based on the shader progs data */
+ job->tmu_dirty_rcl |= v3d->prog.cs->prog_data.vs->base.tmu_dirty_rcl;
+ job->tmu_dirty_rcl |= v3d->prog.vs->prog_data.vs->base.tmu_dirty_rcl;
+ if (v3d->prog.gs_bin) {
+ job->tmu_dirty_rcl |=
+ v3d->prog.gs_bin->prog_data.gs->base.tmu_dirty_rcl;
+ }
+ if (v3d->prog.gs) {
+ job->tmu_dirty_rcl |=
+ v3d->prog.gs->prog_data.gs->base.tmu_dirty_rcl;
+ }
+ job->tmu_dirty_rcl |= v3d->prog.fs->prog_data.fs->base.tmu_dirty_rcl;
+
+ uint32_t num_elements_to_emit = 0;
+ for (int i = 0; i < vtx->num_elements; i++) {
+ struct pipe_vertex_element *elem = &vtx->pipe[i];
+ struct pipe_vertex_buffer *vb =
+ &vertexbuf->vb[elem->vertex_buffer_index];
+ if (vb->buffer.resource)
+ num_elements_to_emit++;
+ }
+
+ uint32_t shader_state_record_length =
+ cl_packet_length(GL_SHADER_STATE_RECORD);
+ if (v3d->prog.gs) {
+ shader_state_record_length +=
+ cl_packet_length(GEOMETRY_SHADER_STATE_RECORD) +
+ cl_packet_length(TESSELLATION_GEOMETRY_COMMON_PARAMS) +
+ 2 * cl_packet_length(TESSELLATION_GEOMETRY_SHADER_PARAMS);
+ }
+
+ /* See GFXH-930 workaround below */
+ uint32_t shader_rec_offset =
+ v3d_cl_ensure_space(&job->indirect,
+ shader_state_record_length +
+ MAX2(num_elements_to_emit, 1) *
+ cl_packet_length(GL_SHADER_STATE_ATTRIBUTE_RECORD),
+ 32);
+
+ /* XXX perf: We should move most of the SHADER_STATE_RECORD setup to
+ * compile time, so that we mostly just have to OR the VS and FS
+ * records together at draw time.
+ */
+
+ struct vpm_config vpm_cfg_bin, vpm_cfg;
+ v3d_compute_vpm_config(&v3d->screen->devinfo,
+ v3d->prog.cs->prog_data.vs,
+ v3d->prog.vs->prog_data.vs,
+ v3d->prog.gs ? v3d->prog.gs_bin->prog_data.gs : NULL,
+ v3d->prog.gs ? v3d->prog.gs->prog_data.gs : NULL,
+ &vpm_cfg_bin,
+ &vpm_cfg);
+
+ if (v3d->prog.gs) {
+ v3d_emit_gs_state_record(v3d->job,
+ v3d->prog.gs_bin, gs_bin_uniforms,
+ v3d->prog.gs, gs_uniforms);
+
+ struct v3d_gs_prog_data *gs = v3d->prog.gs->prog_data.gs;
+ v3d_emit_tes_gs_common_params(v3d->job,
+ gs->out_prim_type,
+ gs->num_invocations);
+
+ /* Bin Tes/Gs params */
+ v3d_emit_tes_gs_shader_params(v3d->job,
+ vpm_cfg_bin.gs_width,
+ vpm_cfg_bin.Gd,
+ vpm_cfg_bin.Gv);
+
+ /* Render Tes/Gs params */
+ v3d_emit_tes_gs_shader_params(v3d->job,
+ vpm_cfg.gs_width,
+ vpm_cfg.Gd,
+ vpm_cfg.Gv);
+ }
+
+ emit_shader_state_record(v3d, job, info, vtx,
+ cs_uniforms, vs_uniforms, fs_uniforms,
+ &vpm_cfg_bin, &vpm_cfg);
bool cs_loaded_any = false;
const bool cs_uses_builtins = v3d->prog.cs->prog_data.vs->uses_iid ||
--
2.39.2
From bb2a9b1ffdb91ad94d1e1f29594d262460d5e29b Mon Sep 17 00:00:00 2001
From: Iago Toral Quiroga <itoral@igalia.com>
Date: Thu, 23 Nov 2023 11:33:57 +0100
Subject: [PATCH 3/3] v3dv: support 2712D0
2712D0 has V3D 7.1.10 which included draw index and
base vertex in the shader state record packet, shuffling
the locations of most of its fields. Handle this at run
time by emitting the appropriate packet based on the
V3D version since our current versioning framework doesn't
support changes based on revision number alone.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29189>
---
src/broadcom/vulkan/v3dvx_cmd_buffer.c | 96 +++++++++++++++++---------
src/broadcom/vulkan/v3dvx_pipeline.c | 75 +++++++++++++++++---
2 files changed, 130 insertions(+), 41 deletions(-)
diff --git a/src/broadcom/vulkan/v3dvx_cmd_buffer.c b/src/broadcom/vulkan/v3dvx_cmd_buffer.c
index 65b18ae639cd..09fae0307bcd 100644
--- a/src/broadcom/vulkan/v3dvx_cmd_buffer.c
+++ b/src/broadcom/vulkan/v3dvx_cmd_buffer.c
@@ -2333,6 +2333,13 @@ v3dX(cmd_buffer_emit_gl_shader_state)(struct v3dv_cmd_buffer *cmd_buffer)
uint32_t shader_state_record_length =
cl_packet_length(GL_SHADER_STATE_RECORD);
+#if V3D_VERSION >= 71
+ if (v3d_device_has_draw_index(&pipeline->device->devinfo)) {
+ shader_state_record_length =
+ cl_packet_length(GL_SHADER_STATE_RECORD_DRAW_INDEX);
+ }
+#endif
+
if (pipeline->has_gs) {
shader_state_record_length +=
cl_packet_length(GEOMETRY_SHADER_STATE_RECORD) +
@@ -2380,39 +2387,64 @@ v3dX(cmd_buffer_emit_gl_shader_state)(struct v3dv_cmd_buffer *cmd_buffer)
pipeline->device->default_attribute_float;
#endif
- cl_emit_with_prepacked(&job->indirect, GL_SHADER_STATE_RECORD,
- pipeline->shader_state_record, shader) {
-
- /* FIXME: we are setting this values here and during the
- * prepacking. This is because both cl_emit_with_prepacked and v3dvx_pack
- * asserts for minimum values of these. It would be good to get
- * v3dvx_pack to assert on the final value if possible
- */
- shader.min_coord_shader_input_segments_required_in_play =
- pipeline->vpm_cfg_bin.As;
- shader.min_vertex_shader_input_segments_required_in_play =
- pipeline->vpm_cfg.As;
-
- shader.coordinate_shader_code_address =
- v3dv_cl_address(assembly_bo, vs_bin_variant->assembly_offset);
- shader.vertex_shader_code_address =
- v3dv_cl_address(assembly_bo, vs_variant->assembly_offset);
- shader.fragment_shader_code_address =
- v3dv_cl_address(assembly_bo, fs_variant->assembly_offset);
-
- shader.coordinate_shader_uniforms_address = cmd_buffer->state.uniforms.vs_bin;
- shader.vertex_shader_uniforms_address = cmd_buffer->state.uniforms.vs;
- shader.fragment_shader_uniforms_address = cmd_buffer->state.uniforms.fs;
-
-#if V3D_VERSION == 42
- shader.address_of_default_attribute_values =
- v3dv_cl_address(default_attribute_values, 0);
+#if V3D_VERSION >= 71
+ if (v3d_device_has_draw_index(&pipeline->device->devinfo)) {
+ cl_emit_with_prepacked(&job->indirect, GL_SHADER_STATE_RECORD_DRAW_INDEX,
+ pipeline->shader_state_record, shader) {
+ shader.min_coord_shader_input_segments_required_in_play =
+ pipeline->vpm_cfg_bin.As;
+ shader.min_vertex_shader_input_segments_required_in_play =
+ pipeline->vpm_cfg.As;
+ shader.coordinate_shader_code_address =
+ v3dv_cl_address(assembly_bo, vs_bin_variant->assembly_offset);
+ shader.vertex_shader_code_address =
+ v3dv_cl_address(assembly_bo, vs_variant->assembly_offset);
+ shader.fragment_shader_code_address =
+ v3dv_cl_address(assembly_bo, fs_variant->assembly_offset);
+ shader.coordinate_shader_uniforms_address = cmd_buffer->state.uniforms.vs_bin;
+ shader.vertex_shader_uniforms_address = cmd_buffer->state.uniforms.vs;
+ shader.fragment_shader_uniforms_address = cmd_buffer->state.uniforms.fs;
+ shader.any_shader_reads_hardware_written_primitive_id =
+ (pipeline->has_gs && prog_data_gs->uses_pid) || prog_data_fs->uses_pid;
+ shader.insert_primitive_id_as_first_varying_to_fragment_shader =
+ !pipeline->has_gs && prog_data_fs->uses_pid;
+ }
+ } else
#endif
-
- shader.any_shader_reads_hardware_written_primitive_id =
- (pipeline->has_gs && prog_data_gs->uses_pid) || prog_data_fs->uses_pid;
- shader.insert_primitive_id_as_first_varying_to_fragment_shader =
- !pipeline->has_gs && prog_data_fs->uses_pid;
+ {
+ cl_emit_with_prepacked(&job->indirect, GL_SHADER_STATE_RECORD,
+ pipeline->shader_state_record, shader) {
+ /* FIXME: we are setting this values here and during the
+ * prepacking. This is because both cl_emit_with_prepacked and v3dvx_pack
+ * asserts for minimum values of these. It would be good to get
+ * v3dvx_pack to assert on the final value if possible
+ */
+ shader.min_coord_shader_input_segments_required_in_play =
+ pipeline->vpm_cfg_bin.As;
+ shader.min_vertex_shader_input_segments_required_in_play =
+ pipeline->vpm_cfg.As;
+
+ shader.coordinate_shader_code_address =
+ v3dv_cl_address(assembly_bo, vs_bin_variant->assembly_offset);
+ shader.vertex_shader_code_address =
+ v3dv_cl_address(assembly_bo, vs_variant->assembly_offset);
+ shader.fragment_shader_code_address =
+ v3dv_cl_address(assembly_bo, fs_variant->assembly_offset);
+
+ shader.coordinate_shader_uniforms_address = cmd_buffer->state.uniforms.vs_bin;
+ shader.vertex_shader_uniforms_address = cmd_buffer->state.uniforms.vs;
+ shader.fragment_shader_uniforms_address = cmd_buffer->state.uniforms.fs;
+
+ #if V3D_VERSION == 42
+ shader.address_of_default_attribute_values =
+ v3dv_cl_address(default_attribute_values, 0);
+ #endif
+
+ shader.any_shader_reads_hardware_written_primitive_id =
+ (pipeline->has_gs && prog_data_gs->uses_pid) || prog_data_fs->uses_pid;
+ shader.insert_primitive_id_as_first_varying_to_fragment_shader =
+ !pipeline->has_gs && prog_data_fs->uses_pid;
+ }
}
/* Upload vertex element attributes (SHADER_STATE_ATTRIBUTE_RECORD) */
diff --git a/src/broadcom/vulkan/v3dvx_pipeline.c b/src/broadcom/vulkan/v3dvx_pipeline.c
index 478fb9a0fa1e..e515286847bd 100644
--- a/src/broadcom/vulkan/v3dvx_pipeline.c
+++ b/src/broadcom/vulkan/v3dvx_pipeline.c
@@ -399,6 +399,10 @@ v3dX(pipeline_pack_state)(struct v3dv_pipeline *pipeline,
static void
pack_shader_state_record(struct v3dv_pipeline *pipeline)
{
+ /* To siplify the code we ignore here GL_SHADER_STATE_RECORD_DRAW_INDEX
+ * used with 2712D0, since we know that has the same size as the regular
+ * version.
+ */
assert(sizeof(pipeline->shader_state_record) >=
cl_packet_length(GL_SHADER_STATE_RECORD));
@@ -411,6 +415,14 @@ pack_shader_state_record(struct v3dv_pipeline *pipeline)
struct v3d_vs_prog_data *prog_data_vs_bin =
pipeline->shared_data->variants[BROADCOM_SHADER_VERTEX_BIN]->prog_data.vs;
+ bool point_size_in_shaded_vertex_data;
+ if (!pipeline->has_gs) {
+ point_size_in_shaded_vertex_data = pipeline->topology == MESA_PRIM_POINTS;
+ } else {
+ struct v3d_gs_prog_data *prog_data_gs =
+ pipeline->shared_data->variants[BROADCOM_SHADER_GEOMETRY]->prog_data.gs;
+ point_size_in_shaded_vertex_data = prog_data_gs->writes_psiz;
+ }
/* Note: we are not packing addresses, as we need the job (see
* cl_pack_emit_reloc). Additionally uniforms can't be filled up at this
@@ -418,17 +430,62 @@ pack_shader_state_record(struct v3dv_pipeline *pipeline)
* pipeline (like viewport), . Would need to be filled later, so we are
* doing a partial prepacking.
*/
+#if V3D_VERSION >= 71
+ /* 2712D0 (V3D 7.1.10) has included draw index and base vertex, shuffling all
+ * the fields in the packet. Since the versioning framework doesn't handle
+ * revision numbers, the XML has a different shader state record packet
+ * including the new fields and we device at run time which packet we need
+ * to emit.
+ */
+ if (v3d_device_has_draw_index(&pipeline->device->devinfo)) {
+ v3dvx_pack(pipeline->shader_state_record, GL_SHADER_STATE_RECORD_DRAW_INDEX, shader) {
+ shader.enable_clipping = true;
+ shader.point_size_in_shaded_vertex_data = point_size_in_shaded_vertex_data;
+ shader.fragment_shader_does_z_writes = prog_data_fs->writes_z;
+ shader.turn_off_early_z_test = prog_data_fs->disable_ez;
+ shader.fragment_shader_uses_real_pixel_centre_w_in_addition_to_centroid_w2 =
+ prog_data_fs->uses_center_w;
+ shader.enable_sample_rate_shading =
+ pipeline->sample_rate_shading ||
+ (pipeline->msaa && prog_data_fs->force_per_sample_msaa);
+ shader.any_shader_reads_hardware_written_primitive_id = false;
+ shader.do_scoreboard_wait_on_first_thread_switch =
+ prog_data_fs->lock_scoreboard_on_first_thrsw;
+ shader.disable_implicit_point_line_varyings =
+ !prog_data_fs->uses_implicit_point_line_varyings;
+ shader.number_of_varyings_in_fragment_shader = prog_data_fs->num_inputs;
+ shader.coordinate_shader_input_vpm_segment_size = prog_data_vs_bin->vpm_input_size;
+ shader.vertex_shader_input_vpm_segment_size = prog_data_vs->vpm_input_size;
+ shader.coordinate_shader_output_vpm_segment_size = prog_data_vs_bin->vpm_output_size;
+ shader.vertex_shader_output_vpm_segment_size = prog_data_vs->vpm_output_size;
+ shader.min_coord_shader_input_segments_required_in_play =
+ pipeline->vpm_cfg_bin.As;
+ shader.min_vertex_shader_input_segments_required_in_play =
+ pipeline->vpm_cfg.As;
+ shader.min_coord_shader_output_segments_required_in_play_in_addition_to_vcm_cache_size =
+ pipeline->vpm_cfg_bin.Ve;
+ shader.min_vertex_shader_output_segments_required_in_play_in_addition_to_vcm_cache_size =
+ pipeline->vpm_cfg.Ve;
+ shader.coordinate_shader_4_way_threadable = prog_data_vs_bin->base.threads == 4;
+ shader.vertex_shader_4_way_threadable = prog_data_vs->base.threads == 4;
+ shader.fragment_shader_4_way_threadable = prog_data_fs->base.threads == 4;
+ shader.coordinate_shader_start_in_final_thread_section = prog_data_vs_bin->base.single_seg;
+ shader.vertex_shader_start_in_final_thread_section = prog_data_vs->base.single_seg;
+ shader.fragment_shader_start_in_final_thread_section = prog_data_fs->base.single_seg;
+ shader.vertex_id_read_by_coordinate_shader = prog_data_vs_bin->uses_vid;
+ shader.base_instance_id_read_by_coordinate_shader = prog_data_vs_bin->uses_biid;
+ shader.instance_id_read_by_coordinate_shader = prog_data_vs_bin->uses_iid;
+ shader.vertex_id_read_by_vertex_shader = prog_data_vs->uses_vid;
+ shader.base_instance_id_read_by_vertex_shader = prog_data_vs->uses_biid;
+ shader.instance_id_read_by_vertex_shader = prog_data_vs->uses_iid;
+ }
+ return;
+ }
+#endif
+
v3dvx_pack(pipeline->shader_state_record, GL_SHADER_STATE_RECORD, shader) {
shader.enable_clipping = true;
-
- if (!pipeline->has_gs) {
- shader.point_size_in_shaded_vertex_data =
- pipeline->topology == MESA_PRIM_POINTS;
- } else {
- struct v3d_gs_prog_data *prog_data_gs =
- pipeline->shared_data->variants[BROADCOM_SHADER_GEOMETRY]->prog_data.gs;
- shader.point_size_in_shaded_vertex_data = prog_data_gs->writes_psiz;
- }
+ shader.point_size_in_shaded_vertex_data = point_size_in_shaded_vertex_data;
/* Must be set if the shader modifies Z, discards, or modifies
* the sample mask. For any of these cases, the fragment
--
2.39.2
View Git Blame Copy Permalink